引入注意力机制和参数优化的TCN短期风电功率预测

TCN Short-term Wind Power Prediction by Introducing Attention Mechanism and Parameter Optimization

为提高短期风电功率预测的准确性,提出一种基于频率注意力机制和粒子群优化的时间卷积网络短期风电功率预测模型。首先,利用灰色关联分析方法计算气象变量与风电功率的关联度,选取特征变量;其次,引入频率注意力机制改进时间卷积网络残差块,赋予通道不同权重,并基于粒子群优化算法优化时间卷积网络超参数,搭建预测模型;最后,以黑龙江某风电场实测数据为例进行仿真分析,实验结果表明,所提方法能够充分提取风电功率序列的时序特征,提高短期风电功率预测精度。

To improve the accuracy of short-term wind power prediction,a temporal convolutional network(TCN)short-term wind power prediction model based on frequency attention mechanism(Fca)and particle swarm optimization(PSO)is proposed.First,the grey relational analysis(GRA)method is employed to calculate the correlation degrees be-tween meteorological variables and wind power,thus selecting the characteristic variables.Second,Fca is introduced to improve the TCN residual block by assigning different weights to channels.In addition,the TCN hyperparameters are optimized based on the PSO algorithm,and a prediction model is constructed.Finally,the actual measured data from one wind farm in Heilongjiang Province is taken as an example,and simulation analysis is conducted.Experimental re-sults demonstrate that the proposed method can fully extract the temporal characteristics of wind power sequence and ef-fectively improve the accuracy of short-term wind power prediction.